@mathismeadows/roamer-device-auth 1.1.1 → 1.1.2

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@mathismeadows/roamer-device-auth",
-  "version": "1.1.1",
+  "version": "1.1.2",
   "private": false,
   "type": "module",
   "description": "AUTH-25: server-mediated device authorization flow + stdio<->HTTP MCP proxy for Roamer MCP. Used by roamer-bridge.sh for every OS and every default browser (no more Safari-specific branching, AUTH-14) — Cloudflare Access's Managed OAuth has no device-code grant of its own, so RoamerMcp's server implements it and this script talks to that instead of Entra directly.",

package/roamer-device-auth.mjs

@@ -34,6 +34,7 @@ const ROAMER_MCP_ORIGIN = new URL(ROAMER_MCP_URL).origin;
 
 const CACHE_DIR = join(homedir(), ".mcp-auth-device");
 const CACHE_FILE = join(CACHE_DIR, "roamer_tokens.json");
+const PENDING_FILE = join(CACHE_DIR, "roamer_pending.json");
 
 // Bumped whenever the cached shape changes meaningfully. A cache written by a prior
 // mechanism (e.g. the retired Entra-direct flow, AUTH-11/14) happens to share field names
@@ -90,6 +91,35 @@ function clearCachedTokens() {
   return writeFile(CACHE_FILE, "{}", { mode: 0o600 }).catch(() => {});
 }
 
+// Claude Code (or any MCP host) may kill this process and respawn a fresh one if it doesn't
+// see a stdio handshake within its own connect timeout — and the interactive device flow
+// (dialog -> browser -> Cloudflare's consent screen -> poll) routinely takes longer than a
+// human can click through inside a short timeout. Without persisting the in-progress flow,
+// every respawn called /oauth/device/start again, which mints a BRAND NEW device_code/
+// user_code — the user would see a different code every single retry, forever, with no
+// path to ever actually finish signing in. Confirmed live 2026-08-21. Persisting the
+// pending flow means a respawned process resumes polling the SAME still-valid code instead.
+async function readPendingFlow() {
+  try {
+    const pending = JSON.parse(await readFile(PENDING_FILE, "utf8"));
+    if (!pending?.device_code || Date.now() > pending.expiresAt) return null;
+    return pending;
+  } catch {
+    return null;
+  }
+}
+
+async function writePendingFlow(device) {
+  await mkdir(CACHE_DIR, { recursive: true, mode: 0o700 });
+  const pending = { ...device, expiresAt: Date.now() + device.expires_in * 1000 };
+  await writeFile(PENDING_FILE, JSON.stringify(pending, null, 2), { mode: 0o600 });
+  return pending;
+}
+
+function clearPendingFlow() {
+  return writeFile(PENDING_FILE, "{}", { mode: 0o600 }).catch(() => {});
+}
+
 function expiresSoon(tokens) {
   // expires_in is optional in the response shape — an absent value means treat the token as
   // long-lived and rely on a real 401 to trigger re-auth rather than guessing a lifetime.
@@ -178,30 +208,54 @@ async function doGetValidTokens(forceRefresh) {
 
   if (forceRefresh) await clearCachedTokens();
 
-  log("Starting sign-in...");
-  const device = await startDeviceFlow();
+  // Resume an already-in-progress flow (from a process this host killed and respawned)
+  // instead of minting a new device_code the user would have to start over for.
+  let device = await readPendingFlow();
+  let resuming = Boolean(device);
+  if (!resuming) {
+    log("Starting sign-in...");
+    device = await startDeviceFlow();
+    await writePendingFlow(device);
+  } else {
+    log("Resuming an already-in-progress sign-in (a prior process was restarted before it finished)...");
+  }
+
   log(`Go to ${device.verification_uri} and enter code: ${device.user_code}`);
   // Fire-and-forget: the dialog is how the human actually sees this on macOS; polling below
-  // must not wait on it being dismissed.
-  showDeviceCodeDialog(device.verification_uri, device.user_code);
-  try {
-    const { default: open } = await import("open");
-    await open(device.verification_uri_complete);
-  } catch {
-    // Best-effort only — the dialog and QR code below already carry the URL and code.
+  // must not wait on it being dismissed. Skip re-showing it on a resume — the human already
+  // saw it (or is mid-flow in the browser) from the process that started this same code.
+  if (!resuming) showDeviceCodeDialog(device.verification_uri, device.user_code);
+  if (!resuming) {
+    try {
+      const { default: open } = await import("open");
+      await open(device.verification_uri_complete);
+    } catch {
+      // Best-effort only — the dialog and QR code below already carry the URL and code.
+    }
   }
   // Terminal-visible for any client where a human sees this process's stderr/log output
   // (Claude Code CLI, VS Code's integrated terminal) — QR codes are a real usability win
   // for CLI auth over typing an 8-character code by hand.
-  qrcode.generate(device.verification_uri_complete, { small: true }, (qr) => {
-    process.stderr.write(`${qr}\n`);
-  });
+  if (!resuming) {
+    qrcode.generate(device.verification_uri_complete, { small: true }, (qr) => {
+      process.stderr.write(`${qr}\n`);
+    });
+  }
 
-  const fresh = await pollDeviceFlow(device.device_code, device.interval ?? 5);
-  tokens = { ...fresh, obtained_at: Date.now() };
-  await writeCachedTokens(tokens);
-  log("Sign-in complete.");
-  return tokens;
+  try {
+    const fresh = await pollDeviceFlow(device.device_code, device.interval ?? 5);
+    tokens = { ...fresh, obtained_at: Date.now() };
+    await writeCachedTokens(tokens);
+    await clearPendingFlow();
+    log("Sign-in complete.");
+    return tokens;
+  } catch (err) {
+    // A hard failure (expired/denied, not just this process being killed) means the pending
+    // code is genuinely dead — clear it so the next attempt starts a real fresh one instead
+    // of retrying a code that will never succeed.
+    await clearPendingFlow();
+    throw err;
+  }
 }
 
 async function main() {